In November 2022, ChatGPT introduced generative artificial intelligence (AI) to the public. The chatbot generates original text using large language models (LLMs). It’s not the only LLM available, but ChatGPT’s popularity highlights its maturity and visibility. It demonstrates how generative AI can be essential in shaping language-based interactions across industries.  

LLM technology has refined machine translation (MT), increasing localization speed, accuracy, and scalability. As it matures, it promises to reduce costs further, improve customer experiences, and streamline workflow management. However, like any technology, generative AI has its limitations. Here is a look at four broad use cases and why localization still requires human linguists despite these advances. 

How Does Generative AI Work?

To understand what is possible with generative AI, we must first understand how it works. Generative AI uses neural networks to analyze patterns from large data sets, and it uses these patterns to generate original text, images, and other media responding to prompts. For example, a model trained using a data set of photographs can generate an entirely new image when given an input prompt, such as a landscape description. 

Four AI Use Cases for Localization 

Generative AI is not developed initially with localization in mind; its ability to generate original outputs makes it a powerful tool for innovation. AI opens up a world of possibilities for localization processes and enables language service providers (LSPs) to create localized content quickly and efficiently. Some use cases are already integral to localization, while others are still experimental and require refinement. 

Build Custom Localization and QA Tools     

LLM tools may complement traditional translation and quality assurance tools to offer sophisticated content analysis and quality assurance. Localization providers could use existing models to build custom tools that assist with tasks such as: 

• Organizing and managing complex file types. For example, an LLM tool could identify and extract images, audio, and video files from a project and organize them into a separate folder structure for easy access and review. 
• Analyzing source content pre-translation to optimize localization. For example, a tool could check for text expansion, suggest reducing the overall word count, and check for untranslatable text elements.
• Integrating glossaries, term bases, and language tools into the “human-in-the-loop” workflow. For example, a tool could identify context-specific terms to be included in glossaries and term bases and identify mistranslations generated by MT.

Generative adversarial networks (GANs) and variational autoencoders (VAEs) can create original images or adapt existing ones to suit the target locale better. For example, localization providers could build a tool that modifies images by changing clothing styles or adjusting colors based on prompts. This would free up time to focus on more complex design projects and reduce costs associated with stock images.          

Enable Content Personalization Across Industries

AI-powered content personalization can supercharge localization efforts by improving engagement, building brand loyalty, and increasing conversions. Companies that invest in personalization technology outsell their competitors by approximately 30%. LSPs that integrate personalization into the localization process will add value for customers that continuously create content.    

Here’s how it works. Algorithms analyze data to identify patterns and preferences and then use that data to deliver personalized emails, push notifications, in-app messages, and other content. For instance, online retailers can use algorithmic personalization to send customers product recommendations based on their purchase history. 

Content personalization isn’t just limited to retail. Streaming services such as Netflix and Hulu use personalization to recommend movies and TV shows to their users based on their viewing history. And in education, personalization technology can create individualized learning paths for students to ensure they receive tailored content that meets their needs.  

As more industries adopt personalization, advanced AI-powered localization tools will make it easier for LSPs to provide continuous translation at scale. 

Integrate Speech Recognition into the Translation Process

Speech recognition tools help streamline localization by accurately transcribing and translating spoken content. These tools use sophisticated algorithms and deep neural networks to analyze audio data, detect patterns, and convert them into written language. This enables localization providers to: 

• Transcribe video and audio content into text. Allows you to prepare translations for subtitles. For instance, you can transcribe long audio recordings of interviews, convert them into text, and use it to generate subtitles for video content.
• Translate spoken content into the target language. This is useful for live interpretation or pre-translation of pre-recorded content. For example, you can generate real-time translations of conversations between two people in different languages and then generate subtitles in both languages.
• Enable voice user interfaces (VUI). The VUI can understand and interpret the user’s spoken commands and respond in the user’s language. For instance, AI-powered VUIs can interact with devices such as smart speakers and home appliances. 
• Customer support. In multilingual customer support, speech recognition tools assist in identifying and transcribing customer inquiries and directing them to a representative who speaks the appropriate language. For example, a customer service team can quickly identify a customer’s language and route the call to the appropriate representative without human intervention.

In all likelihood, speech recognition tools will become increasingly important as the volume of spoken content grows and customer support moves online.     

Translate Lesser-known Languages at Scale 

Today, content is translated mainly into about 20 world languages. Yet as internet access increases in emerging markets, people want content in their language. At the same time, the low number of local linguists makes it challenging to translate less common languages at scale. LLMs can help LSPs tackle this challenge by leveraging unsupervised learning techniques. 

Unsupervised learning involves training an LLM model on a large amount of unannotated data without direct supervision or labels. This approach is helpful for low-resource languages or languages with limited annotated data. For instance, unsupervised learning methods have been used to train LLMs in languages with limited lexicons, such as West African languages.

However, it’s critical to note that models for low-resource languages will have limited performance compared to more widely spoken languages. That means it may not be suitable for content that requires a high degree of accuracy or cultural sensitivity. 

The Limitations of Generative AI

Although generative AI has made significant progress in recent years, it still has limitations. As a result, human linguists need to review and modify the output. 

Here are four things to consider before integrating AI into the localization process. 

• Limited creativity. Generative AI produces content similar to what it learned from training data but needs help to create original or creative content. That is because the patterns limit outputs along with the rules it learns from the data. While it works well for product descriptions, technical specifications, and user-generated content, AI isn’t suitable for creative translations such as marketing and advertising. 

• Bias and inaccuracies. Data bias can affect AI models, which occurs when training data isn’t representative of the real world, leading to a biased or inaccurate output. Linguists with expertise should review AI-generated content for racist, sexist, and culturally inappropriate language.  

• Language variables. AI applications need large volumes of data to be trained accurately; most data sets are from significant languages. Consequently, LLMs have lower accuracy rates for less common languages and alternative dialects. To address these language variables, you need a diverse team to ensure accuracy, inclusiveness, and cultural understanding.  
• High computational requirements. Generative AI models are often computationally intensive and require significant resources to run and train, and this can make them impractical and costly for some applications. That means some models must be more mature to create custom localization tools.  

Remember that generative AI works best for generating large volumes of localized content with minimal cost and effort. For example, generative AI can create thousands of localized versions of product descriptions for an eCommerce website. However, projects that require a nuanced understanding of the target language and culture are best left to human translators.    

The Future of AI for Localization 

AI is a rapidly evolving sector with seemingly endless possibilities. Open-source or commercial tools will likely offer more opportunities for localization providers to integrate this technology into workflows. Further developments will significantly expand the possible use cases and drive custom training for specific tasks and outcomes. However, despite these advances, human linguists will continue to play a role by leveraging creativity and cultural expertise.